In current communication transmission, more attention is paid to the quality of voice or audio, and therefore encoding and decoding of a voice signal or an audio signal is becoming a more important procedure in voice or audio signal processing.
In a signal encoding process, in order to improve encoding efficiency, an encoder end generally expects to use as few coded bits as possible to represent a signal to be transmitted. For example, during low-rate encoding, the encoder end usually does not perform encoding on all bands. Considering a feature that human ears are more sensitive to a low-frequency part than to a high-frequency part in a voice signal or an audio signal, generally, more bits are allocated to the low-frequency part for encoding, while only a few bits are allocated to the high-frequency part for encoding; in some cases, the high-frequency part is even not encoded. Therefore, during decoding on a decoder end, a band on which encoding is not performed needs to be restored by means of a blind bandwidth expansion technology.
At present, the decoder end usually uses a time-domain bandwidth extension manner to restore the band on which encoding is not performed. However, in this manner, an extension effect of a voice signal is poor, and an audio signal cannot be processed, and consequently an output voice or audio signal has poor performance.